Mailing-List: contact dev-help@madlib.incubator.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@madlib.incubator.apache.org
From: orhankislal <git@git.apache.org>
To: dev@madlib.incubator.apache.org
Reply-To: dev@madlib.incubator.apache.org
References: <git-pr-112-incubator-madlib@git.apache.org>
In-Reply-To: <git-pr-112-incubator-madlib@git.apache.org>
Subject: [GitHub] incubator-madlib pull request #112: Feautre: Add grouping support for PageRa...
Content-Type: text/plain
Message-Id: <20170412000647.7C971DF97E@git1-us-west.apache.org>
Date: Wed, 12 Apr 2017 00:06:47 +0000 (UTC)
archived-at: Wed, 12 Apr 2017 00:06:52 -0000

Github user orhankislal commented on a diff in the pull request:

    https://github.com/apache/incubator-madlib/pull/112#discussion_r111038142
  
    --- Diff: src/ports/postgres/modules/graph/pagerank.py_in ---
    @@ -158,44 +313,198 @@ def pagerank(schema_madlib, vertex_table, vertex_id, edge_table, edge_args,
             # https://en.wikipedia.org/wiki/PageRank#Damping_factor
     
             # The query below computes the PageRank of each node using the above formula.
    +        # A small explanatory note on ignore_group_clause:
    +        # This is used only when grouping is set. This essentially will have
    +        # the condition that will help skip the PageRank computation on groups
    +        # that have converged.
             plpy.execute("""
                     CREATE TABLE {message} AS
    -                SELECT {edge_temp_table}.{dest} AS {vertex_id},
    -                        SUM({v1}.pagerank/{out_cnts}.{out_cnts_cnt})*{damping_factor}+{random_prob} AS pagerank
    +                SELECT {grouping_cols_select} {edge_temp_table}.{dest} AS {vertex_id},
    +                        SUM({v1}.pagerank/{out_cnts}.{out_cnts_cnt})*{damping_factor}+{random_jump_prob} AS pagerank
                     FROM {edge_temp_table}
    -                    INNER JOIN {cur} ON {edge_temp_table}.{dest}={cur}.{vertex_id}
    -                    INNER JOIN {out_cnts} ON {out_cnts}.{vertex_id}={edge_temp_table}.{src}
    -                    INNER JOIN {cur} AS {v1} ON {v1}.{vertex_id}={edge_temp_table}.{src}
    -                GROUP BY {edge_temp_table}.{dest}
    -            """.format(**locals()))
    +                    INNER JOIN {cur} ON {cur_join_clause}
    +                    INNER JOIN {out_cnts} ON {out_cnts_join_clause}
    +                    INNER JOIN {cur} AS {v1} ON {v1_join_clause}
    +                    {vertices_per_group_inner_join}
    +                {ignore_group_clause}
    +                GROUP BY {grouping_cols_select} {edge_temp_table}.{dest}
    +            """.format(grouping_cols_select=edge_grouping_cols_select+', '
    +                    if grouping_cols else '',
    +                random_jump_prob='MIN({vpg}.{random_prob})'.format(**locals())
    +                    if grouping_cols else random_probability,
    +                vertices_per_group_inner_join="""INNER JOIN {vertices_per_group}
    +                    AS {vpg} ON {vpg_join_clause}""".format(**locals())
    +                    if grouping_cols else '',
    +                ignore_group_clause=' WHERE '+get_ignore_groups(
    +                    summary_table, edge_temp_table, grouping_cols_list)
    +                    if iteration_num>0 and grouping_cols else '',
    +                **locals()))
             # If there are nodes that have no incoming edges, they are not captured in the message table.
             # Insert entries for such nodes, with random_prob.
             plpy.execute("""
                     INSERT INTO {message}
    -                SELECT {vertex_id}, {random_prob}::DOUBLE PRECISION AS pagerank
    -                FROM {cur}
    -                WHERE {vertex_id} NOT IN (
    +                SELECT {grouping_cols_select} {cur}.{vertex_id}, {random_jump_prob} AS pagerank
    +                FROM {cur} {vpg_from_clause}
    +                WHERE {vpg_where_clause} {vertex_id} NOT IN (
    --- End diff --
    
    The select clause gives you every vertex that exists in the message table regardless of its group. Assume message has 1,2,3 for group 'a' and 1,2,3,4 for group 'b'. This check will remove 4 for group 'a' as well. You can use NOT EXISTS clause to check for multiple columns simultaneously.


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---